ES2847976T3 - Procedure to determine a rapid intravenous injection to be administered using an insulin delivery device, a computer program product, and a data processing device - Google Patents

Abstract

A method of determining a rapid intravenous injection to be administered by an insulin delivery device, comprising, in a data processing device having one or more processors, - receiving glucose measurement data indicative of measurement signals the level of glucose detected by a continuous glucose sensor for a person; and - determining, by the one or more processors at a point in time, for the person, a rapid intravenous injection of insulin to be administered by the insulin delivery device, determining the rapid intravenous injection of insulin based on a functional relationship provided by a formula comprising formula expressions indicative of - a glucose level derived from glucose measurement data; - a ratio of carbohydrates to insulin (CIR); and - an insulin sensitivity factor (ISF); wherein the determination further comprises at least one of, - for the carbohydrate to insulin ratio, determining a current CIR value based on CIR data indicative of the carbohydrate to insulin ratio for at least one of - a period of time from the start of a previous meal to before the start of a current meal on a current day comprising the point in time; - current food; and - the previous meal; and - for the insulin sensitivity factor, determining a current ISF value based on ISF data indicative of the insulin sensitivity factor for at least one of - the period of time from the start of the previous meal to before from the start of the current meal on a current day comprising the point in time; - current food; and - the previous meal; wherein the determination of the rapid intravenous injection of insulin is carried out by a data processing device provided in a system selected from the following group: an insulin pen; an insulin pump; a remote control device configured to remotely control an insulin pump or insulin pen; a continuous glucose meter; a controller for a continuous glucose meter; a blood glucose meter; and a portable device such as a mobile phone, a tablet or a smart watch, characterized in that the determination of the rapid intravenous injection of insulin further comprises at least one of - determining a rapid intravenous injection of insulin for a lunch after a breakfast previously taken on the current day and - determine a rapid intravenous injection of insulin for dinner after a previous lunch taken on the current day, in which, for lunch, the current CIR value ** (See formula) ** and the current ISF value ** (See formula) ** are determined as follows: ** (See formula) ** where - ** (See formula) ** are, for an ABC adapted rapid intravenous injection calculation , the carbohydrate to insulin ratio and insulin sensitivity values for lunch derived from an average daytime glucose level profile for the person; - ** (See formula) ** and ** (See formula) ** are the carbohydrate to insulin ratio and insulin sensitivity values for breakfast derived from the average daytime glucose level profile for the person ; - ** (See formula) ** and ** (See formula) ** are the carbohydrate to insulin ratio and insulin sensitivity values determined from a postprandial glucose trajectory with respect to breakfast of the previous breakfast eaten earlier in the current day for the person; and - the factor f is an adjustment parameter; and where, for dinner, the current CIR value ** (See formula) ** and the current ISF value ** (See formula) ** are determined as follows: ** (See formula) ** in which - ** (See formula) ** and ** (See formula) ** are, for the ABC adapted rapid intravenous injection calculation, the carbohydrate-to-insulin ratio and insulin sensitivity values for the dinner derived from the person's average daytime glucose level profile; - ** (See formula) ** and ** (See formula) ** are the carbohydrate to insulin ratio and insulin sensitivity values for lunch derived from the average daytime glucose profile for the person ; - ** (See formula) ** and ** (See formula) ** are the values of the ratio of carbohydrates to insulin and insulin sensitivity determined from, with respect to the lunch meal, a trajectory postprandial glucose from the previous lunch eaten earlier in the current day for the person; y - the factor f is a fitting parameter,

Description

DESCRIPTION

Procedure to determine a rapid intravenous injection to be administered using an insulin delivery device, a computer program product, and a data processing device

The present invention relates to a method for determining a rapid intravenous injection to be administered by an insulin delivery device, a computer program product, and a data processing device.

Background

Diabetes is a global epidemic that affected approximately 8.8% of the world's population in 2015 and it is estimated that 5 million diabetes-related deaths that year. Estimated global health spending for diabetes in 2015 was $ 673 billion, which corresponds to approximately 12% of the global budget. Most of those costs are for the treatment of diabetes-related complications. These complications often occur due to incorrect dosing of diabetes medication. Both chronically elevated BG (hyperglycemia) due to underdosing and dangerously low BG (hypoglycemia) due to overdose should be avoided. Due to the high risk of hypoglycemia, the correct dosage of insulin in type 1 diabetes mellitus (DM1) is an especially complex task. The immense interpatient and intrapatient variability in insulin requirements requires daily adaptation of treatment doses. For each meal, an adequate dose of rapid-acting insulin should be injected to counteract the effect of the carbohydrates in the meal on the BG level (also called “prandial rapid intravenous injection”). To determine prandial insulin doses, many patients rely on so-called rapid intravenous (BC) calculators. These are procedures that determine the amount of insulin required based on BG level, carbohydrates ingested at meals, as well as patient-specific BC and time of day settings. Specifically, values may be required for the carbohydrate to insulin ratio (CIR) and the insulin sensitivity factor (ISF). In addition, the duration of insulin action (DIA) can be considered. Choosing those values correctly is difficult.

Florian Reiterer et al. in "Performance assessment of estimation methods for CIR / ISF in bolus calculators" they evaluate the procedures for determining the CIR and ISF estimates by comparing the CIR and ISF estimates with the patient-specific and time-of-day values used by the patients of a clinical trial. Procedures based on detailed data analysis were found to work especially well for predicting CIR, while simple rules of thumb tend to give better results for predicting ISF.

Florian Reiterer et al. in "Identification of diurnal patterns in insulin action from measured CGM data for patients with T1DM" they propose a procedure to identify diurnal changes in the action of insulin in patients suffering from type 1 diabetes mellitus (DM1) based on data registered by systems continuous glucose monitoring (CGMS). The data is fit using a continuous time transfer function that includes time-dependent terms. The values identified for insulin requirements per gram of carbohydrate are compared to the patient-specific carbohydrate-to-insulin ratios used for the calculation of rapid intravenous insulin requirements.

EP 1571 582 A2 discloses an apparatus and a method for determining the ratio of carbohydrates to insulin (CIR) and the insulin sensitivity factor (ISF) of a patient, in which these determined values, i.e. Estimates, along with values for current blood glucose and the deviation from target blood glucose, are used to determine a rapid intravenous injection of insulin in light of carbohydrate intake over a particular time period.

US 2012/02226259 A1 discloses a device, a system and a method for adjusting fluid management based on past or historical fluid management data and / or personal parameters of a user. The devices, and the corresponding systems and procedures, comprise a dispensing unit configured to deliver a fluid from a reservoir to the body of a user and a processor that has instructions operating therein to retrieve data related to one or more time windows. of a memory, evaluate a correction delivery for the one or more time windows based on the data, determine a new value of CIR (ratio of carbohydrates to insulin) for the one or more time windows if the administration of correction is performed regularly after a prandial rapid intravenous injection and / or determining a new basal administration profile for the one or more time windows if the correction administration regularly precedes a prandial rapid intravenous injection.

US 2015/0217055 A1 discloses an insulin administration method that includes receiving blood glucose measurements from a patient on a data processing device of a glucometer. Blood glucose measurements will be separated by a time interval. The procedure also includes receiving information from the patient on the data processing device and selecting an insulin treatment. subcutaneous from a collection of subcutaneous insulin treatments. Selection is based on blood glucose measurements and patient information. Selection includes one or more of a standard subcutaneous program, a subcutaneous program without prandial rapid intravenous injections, a subcutaneous meal-by-meal program without carbohydrate counting, a subcutaneous meal-by-meal program with carbohydrate counting, and a subcutaneous program for non-diabetic patients. . The procedure also includes executing, using the data processing device, the selected subcutaneous insulin treatment.

Summary

An object of the invention is to provide an improved technology for determining a rapid intravenous injection of insulin to be administered by an insulin delivery device, wherein the rapid intravenous injection of insulin can be more reliably and accurately determined for a person.

To meet the objective, a method is provided for determining a rapid intravenous injection to be administered by an insulin delivery device according to independent claim 1. In addition, a computer program product and an insulin processing device are provided. Data according to claims 6 and 7. Further embodiments are provided in the dependent claims.

According to one aspect, a method is provided for determining a rapid intravenous injection to be administered by an insulin delivery device, the method comprising, in a data processing device having one or more processors: receiving data from glucose measurement indicative of glucose level measurement signals detected by a continuous glucose sensor for a person; and determining by the one or more processors at a point in time, for the person, a rapid intravenous injection of insulin to be administered by the insulin delivery device. Rapid intravenous insulin injection is determined based on a functional relationship provided by a formula comprising formula expressions indicative of a glucose level derived from glucose measurement data; a carbohydrate to insulin ratio (CIR); and an insulin sensitivity factor (ISF). The determination further comprises at least one of the following: (i) for the carbohydrate to insulin ratio, determining a current CIR value based on CIR data indicative of the carbohydrate to insulin ratio for at least one of a period of time from the start of a previous meal to before the start of a current meal on a current day comprising the point in time; current food; and the previous meal; and, (ii) for the insulin sensitivity factor, determining a current ISF value based on ISF data indicative of the insulin sensitivity factor for at least one of the time period from the start of the previous meal to before the start of the current meal on a current day comprising the point in time; current food; and the previous meal. The determination of the rapid intravenous injection of insulin is carried out by a data processing device provided in a system selected from the following group: an insulin pen; an insulin pump; a remote control device configured to remotely control an insulin pump or insulin pen; a continuous glucose meter; a controller for a continuous glucose meter; a blood glucose meter; and a portable device such as a mobile phone, a tablet, or a smart watch.

In accordance with yet another aspect, a data processing device is provided, the device comprising one or more processors configured to determine a rapid intravenous injection to be administered by an insulin delivery device by the method described above. In addition, a computer program product is provided.

Patient-specific factors or formal expressions (which may also be referred to more generally as user-specific or custom) apply to both the carbohydrate-to-insulin ratio and the insulin sensitivity factor. For at least one of said factors, a current or new value is determined and applied to determine the rapid intravenous injection of insulin for a user. These new values can substitute a previously provided and potentially applied historical or previous factor for the calculation of the previous rapid intravenous injection.

The insulin delivery device can be an insulin pump and an insulin pen.

With respect to the start of the previous meal (start point in time), the onset may be within one of the following time periods: time period of up to 15 min before ingesting the first meal of the previous meal, from alternatively up to 10 min, up to 5 min or up to 1 min before ingesting the first meal of the previous meal. With respect to the start of the current meal (start point in time), the onset may be within one of the following time periods: period up to 15 min before but not including the ingestion of the first meal of the current meal, alternatively up to 10 min, up to 5 min or up to 1 min before but not including the ingestion of the first meal of the current meal.

Determining the rapid intravenous injection of insulin further comprises determining a rapid intravenous injection of insulin for lunch after a previous breakfast eaten on the current day; and Determine, for lunch, the current CIR value ( CIR ^{^ ^ a} l ²⁰ ) and the current ISP value (/ SF “C¡J £“ f ^zo ) as follows:

in which

- CIR ^{£ jJ¡ lu erzo} and ISF ^{£ g ™ ugh} are, for an adapted rapid intravenous injection (AUC) calculation, the carbohydrate-to-insulin ratio and insulin sensitivity values for lunch derived from a profile average daytime glucose level for the person;

- ClR ^{^ l} Sg ^{V one} and lSF ^{^ ggayuno} are the carbohydrate to insulin ratio and insulin sensitivity values for breakfast derived from the person's average daytime glucose level profile; - ClR ^{g and ^ g ^ one} and lSFfyS ^{s [g and one} are the values of the ratio of carbohydrates to insulin and insulin sensitivity determined, with respect to breakfast, from a postprandial glucose trajectory of the previous breakfast eaten plus early in the current day for the person; and

- the factor f is an adjustment parameter.

In one embodiment, the factor f can be a value in the range from 0 to 1. The factor f is used as a fractional threshold to limit the parameter change.

The above values or factors can be determined from the glucose measurement data provided. Alternatively, one or more of the values or factors may be provided to the data processing device directly without the data processing device having or determining them. Determining the rapid intravenous injection of insulin further comprises determining a rapid intravenous injection of insulin for a dinner following a previous meal ingested on the current day; and determine, for dinner, the current CIR value ( CIRaltuai) and the current ISF value ( SF! £ ™ al) as follows:

in which

- CIRjigc1 and ISFJÍb ™ are, for the adapted rapid intravenous injection (AUC) calculation, the carbohydrate to insulin ratio and insulin sensitivity values for dinner derived from the average daytime glucose level profile for the person ;

- LUNCH CIR and LUNCH SFAg are the carbohydrate-to-insulin ratio and insulin sensitivity values for lunch derived from the person's average daytime glucose profile; - CIRsy ™ i¿erzo and ISFsy ™ ¿erzo are the carbohydrate-to-insulin ratio and insulin sensitivity values determined, relative to the lunch meal, from a postprandial glucose trajectory of the previous lunch eaten earlier in the current day for the person; and

- the factor f is an adjustment parameter.

The above values or factors can be determined from the glucose measurement data provided. Alternatively, one or more of the values or factors may be provided to the data processing device directly without the data processing device having or determining them. The determination may further comprise determining the rapid intravenous injection of insulin based on a functional relationship provided by a formula comprising indicative formula expressions of the total amount of insulin administered within a first time interval and the total amount of carbohydrates ingested by the person of interest within a second time interval.

The method may further comprise at least one of the following: sending rapid intravenous injection data indicative of the rapid intravenous injection of insulin to be delivered by the insulin delivery device through an output device connected to the processing device of data; and transmitting the output data to the insulin delivery device. For example, the output data may comprise at least one of video and audio data indicative of the rapid intravenous injection of insulin to be delivered by the insulin delivery device. Video data can be sent through a display device. Alternatively or in addition, the output data may comprise control data configured to control the operation of the insulin delivery device to deliver the determined rapid intravenous injection of insulin.

Rapid intravenous injection of insulin ( rapid intravenous injection), for the person, can be determined based on the following formula:

CHO G r

Rapid intravenous injection _{preu ob jet iv or}

+ - IOB

CIR Tsf

where CHO is the carbohydrate content of a meal in grams, Gpre is the preprandial glucose level, Goggle is a target glucose level for the postprandial glucose level, ISF is the insulin sensitivity value or factor, CIR is the value of carbohydrates with respect to insulin and IOB is insulin on board. The glucose level, for the person, can be provided as follows: Gpre is the preprandial glucose level and Gog is a target glucose level for the postprandial glucose level.

With respect to the computer program product and the data processing device, the disclosed embodiments for the above method can be applied mutatis mutandis.

Description of other modes of implementation

In the following, embodiments are described, by way of example, with reference to the figures. In the figures they show:

Fig. 1 a graphical representation of the measured glucose data and the parameters identified from the model for a patient (patient 0013) for day 2 and FIT = 67.87%;

Fig. 2 a graphical representation of the measured glucose data and the parameters identified from the model for the patient for day 3 and FIT = 69.93%;

Fig. 3 a graphical representation of the measured glucose data and the parameters identified from the model for the patient for day 4 and FIT = 63.81%;

FIG. 4 a graphical representation of the measured glucose data and the parameters identified from the model for the patient for day 5 and FIT = 64.51%;

Fig. 5 a graphical representation of the measured glucose data and the parameters identified from the model for the patient for day 6 and FIT = 66.68%;

Fig. 6 a graphical representation of the measured glucose data and the parameters identified from the model for the patient for day 7 and FIT = 66.16%;

Fig. 7 a graphical representation for a comparison of the identified K2 / K1 profiles and the clinical CIR values used for the patient;

Fig. 8 a graphical representation for comparison of the identified K2 profiles and the clinical ISF values used for the patient;

Fig. 9 a graphical representation for a comparison of K2 / K1 and the clinical CIR values used for another patient (patient 0033);

Fig. 10 a graphical representation for a comparison of K2 / K1 and the clinical CIR values used for still another patient (patient 0051);

Fig. 11a a graphical representation for a comparison of the calculated CIR (left) and ISF (right) values versus the clinical values used (all meals combined);

Fig. 11b a graphical representation for a comparison (box plots) of the calculated CIR and ISF values divided by the clinical values used (all meals combined);

Fig. 12 a graphical representation for a comparison of the calculated and clinical CIR and ISF values used for breakfast;

Fig. 13 a graphical representation for a comparison of the calculated and clinical CIR and ISF values used for lunch;

Fig. 14 a graphical representation for a comparison of the calculated and clinical CIR and ISF values used for dinner;

FIG. 15 a graphical representation of time in hypoglycemia as a function of day from a virtual test;

FIG. 16 a graphical representation of time in hyperglycemia as a function of day from a virtual test;

Fig. 17 a graphical representation of the value of the cost function V (W = 5) as a function of the day of the virtual test;

Fig. 18 a graphical representation of the CIR correlation between the identified factors from different meals on the same day;

Fig. 19 a graphical representation of the ISF correlation between the identified factors from different meals on the same day;

Fig. 20 a graphical representation of an effect of including last meal information for the calculation of the next rapid intravenous injection: time in hypoglycemia and hyperglycemia as a function of factor f; and

Fig. 21 a graphical representation of an effect of including information from the last meal for the calculation of the next rapid intravenous injection: value of the cost function V (W = 5) as a function of the factor f.

An essential aspect of insulin treatment in DM1 is to dose insulin in such a way that dangerous hypoglycemic episodes are avoided, but at the same time prolonged hyperglycemic episodes are also minimized. The most common way to dose insulin in DM1 is the so-called basal rapid intravenous injection therapy. Patients or, generically called, users continuously supply their body with a small amount of insulin that is intended to keep their BG (blood glucose) level more or less constant in the absence of major disturbances. This amount of insulin is known as basal insulin and normally represents 30 to 60% of the total daily insulin needs.

Patients receiving multiple daily injections (MDI) treatment, that is, patients using an insulin pen or syringe to dose insulin, typically use one or two long-acting insulin injections daily for this basal purpose. After a couple of days with a constant basal insulin injection schedule, the concentration of the corresponding analog in the bloodstream is more or less constant and, as a result, the slow-acting insulin is metabolized at a more or less constant rate.

Patients using an insulin pump (CSII treatment), on the other hand, use a rapid-acting insulin analog to deliver the required basal insulin that is infused according to a preprogrammed profile as a function of the time of day. The basal insulin profile as a function of the time of day is called the basal rate. In an insulin pump, the amount of basal insulin as a function of the time of day can be adjusted much more precisely than with slow-acting insulin injections, and can be varied or even interrupted depending on physiological needs such as stress or exercise.

In basal rapid intravenous injection therapy, patients with DM1 must also deliver so-called rapid intravenous insulin injections to their body. While basal insulin is designed to keep the BG more or less constant in the absence of major challenges, rapid intravenous injections are used to counteract the effect of those challenges on the BG level, as well as to correct the BG level in case it is above the target area (by injecting a so-called “correction rapid intravenous injection”).

Meals can be a major challenge resulting in a rapid increase in BG. Normally, it is assumed that the amount of rapid intravenous insulin required for a meal (the so-called "prandial rapid intravenous injection") is proportional to the amount of carbohydrate ingested. Therefore, it is essential for the patient to estimate the carbohydrate content of meals, which can be achieved by counting basic or advanced carbohydrates or based solely on the patient's experience and knowledge of their own habits. The essence of the basic carbohydrate count is that the patient understand the relationship between food intake, physical activity and insulin injections and BG level in a qualitative sense.

People using the basic carbohydrate count are encouraged to eat a consistent amount of carbohydrates at similar times each day. The insulin dose is then adjusted for that “standard day”.

In advanced carbohydrate counting (ACC), on the other hand, the insulin dose at mealtime is individually adjusted for each meal based on its carbohydrate content. This offers more flexibility, but is associated with greater effort for the patient. The estimated carbohydrate content of meals is used in the ACC in conjunction with patient-specific multiplication factors and information on the target preprandial and postprandial BG level to determine the correct needs for rapid intravenous insulin injection.

In ACC, the patient estimates the carbohydrate content of the meal and injects an amount of insulin as a prandial rapid intravenous injection that is directly proportional to the amount of carbohydrates. Determining the exact carbohydrate content of meals is difficult and requires a lot of experience. In addition to an estimate of the carbohydrate content of the meal, a measurement of the actual preprandial BG level (usually determined with a BG meter) should also be available. Using the carbohydrate estimate and the preprandial b G level, the required amount of insulin in rapid intravenous injection can be determined using the following formula (Warshaw et al., American Diabetes Association, 2008; Walsh et al., 5.1. San Diego , CA: Torrey Pines Press, 2013):

In formula (1), “Rapid intravenous injection” corresponds to the insulin needs in rapid intravenous injection, CHO is the carbohydrate content of the meal in grams, BGpre is the preprandial BG, Target BG describes the target value for the postprandial BG and IOB stands for insulin on board. The formula for calculating the amount of rapid intravenous insulin required comprises three terms:

- The first formula expression or term in formula (1) is used to counteract the effect of food intake on BG. This term is usually significantly larger than the other two terms. The proportionality factor CIR in this term is the so-called ratio of carbohydrates to insulin, that is, the amount of carbohydrates that are counteracted by 1 IU (unit of insulin).

- The second formula expression is used for BG corrections. The amount of insulin injected from this term is proportional to the difference between BGpre and BGtarget. The ISF factor, the insulin sensitivity factor, describes how many mg / dl the BG level will decrease per 1 IU injected.

- The third formula expression, IOB, means insulin on board, that is, insulin in rapid intravenous injection from previous injections that is still active in the body.

Formula (1) can be used to calculate the required rapid intravenous injection of insulin.

The third term in formula (1), IOB, is used to reduce the risk of overdosing insulin. Even rapid-acting insulin takes several hours to complete its full action (typically around 5 to 6 hours). To determine the IOB, the PK / PD profile of the injected insulin analog must be known. The PK / PD profile refers to the pharmacokinetic and pharmacodynamic (PK / PD) profile of action of the most common available insulins and insulin analogs.

Automatic BCs can have an average insulin action curve stored. However, the time required for the full action of insulin in rapid intravenous injection, known as DIA, must be entered in the automatic BC by the patient or the doctor (depending on the type of insulin analog, but also on the characteristics of the patient). The DIA value is then used to scale the average insulin action curve. More information on DIA can be found in, for example, Walsh et al. (Journal of Diabetes Science and Technology, vol. 8, no. 1, pp. 170-178, 2014.).

The CIR and ISF factors or values can be tailored specifically to the patient to allow good control of the BG level. Due to the enormous effect of interpatient and intrapatient variability, this is difficult. To obtain a first estimate, simple correlations are often used that link CIR and ISF to patient-specific quantities that are easy to assess.

Clinical trial data

To test the procedures described here, data from two clinical trials have been used, both conducted at the Institute of Diabetes Technology, Ulm, Germany. Detailed protocols for those two trials can be found in (Zschornack et al., Journal of Diabetes Science and Technology, vol. 7, no. 4, pp.

815-823, 2013) and (Freckmann et al., Journal of Diabetes Science and Technology, vol. 7, no. 4, pp. 842-853, 2013). The main objective of these trials was to test the measurement performance of CGM devices. The clinical protocols of both studies have been designed in accordance with the recommendations of the guideline POCT05-A (Clinical and Laboratory Standards Institute, document CLSI POCT05-A, Wayne, PA, 2008).

In the following, a methodology for calculating estimates of continuous glucose traces from CGM (continuous glucose monitoring) and SMBG (self-monitoring of blood glucose) data is disclosed. Additionally, in this methodology the time delay between BG and the signal measured by the CGM is estimated. The CGM time delays are analyzed with respect to their patient-specific component and correlations are sought between the CGM time delays and other patient-specific quantities. Based on these findings, a strategy is proposed to estimate therapeutic settings for insulin treatment in DM1 based on the identified CGM time delays.

Identification of CIR / ISF from glucose dynamics

As a basis for identifying the CIR and ISF values from the measured glucose dynamics, the following third-order transfer function is chosen that includes an integrating term to model the response of BG to carbohydrate intake, as well as to rapid intravenous insulin injections:

K i

BG ( s) ^{* i}

(1 sT1) 2s ^{• D (s )} • U ( s)

( 1 + s T 2 and s

(2)

In this formula, BG ( s) describes the level of BG, D ( s) the carbohydrates from the food intake and U ( s) the rapid intravenous insulin injections, all in the Laplace domain. Other influencing factors, such as stress, sports, composition of mixed meals, etc., are not incorporated into this model structure. s refers to the Laplace complex frequency.

This model has been proposed and applied as such (see, for example, Kirchsteiger et al., In Proceedings of the 49th IEEE conference on decision and control (CDC), 2011, pp. 5176-5181; Kirchsteiger et al., International Journal of Control, vol. 87, no. 7, pp. 1454-1466, 2014). The advantage of using the model structure formula (2) is that the model parameters have a clear physiological meaning (see also Kirchsteiger et al., In Proceedings of the 2014 American control conference (acc), June 2014, pp. 5465-5470). The constant K1 describes the effect of 1 gram of carbohydrates on BG, while K2 predicts the effect of 1 IU of insulin in rapid intravenous injection (both for t ^ ~). The time constants T1 and T2 are proportional to the response time of the BG to carbohydrate and insulin inputs. Interestingly, the constant -K2 has the same physiological meaning as the ISF factor in the BC formula (1). Furthermore, the physiological interpretation of the mixed constant (- K2 / K1) is identical to that of CIR. As implicitly assumed using the BC formula (1) to calculate insulin needs in rapid intravenous injection, carbohydrate and insulin inputs have a persistent effect on BG in the model structure (2), which is due to the fact that both transfer functions contain an integrating term (pole at s = 0). In this model structure, only rapid intravenous insulin is treated as input and thus has an effect on glucose dynamics. Basal insulin is assumed to be well adjusted and therefore would maintain BG at a constant level in the absence of challenges to glucose metabolism.

The use of such gray box model structures is known to identify CIR and ISF values. The first attempt to do so is reported in (Percival et al., Journal of Diabetes Science and Technology, vol. 4, no. 5, pp.

1214-1228, 2010), however, with only limited success. The structure of the model (formula (2)) was first reported in (Kirchsteiger et al., In Proceedings of the 18th IFAC world congress, Milan, Italy, August 2011, pp. 11761-11766) as a tool to identify interval models. The same structure is then used in a computer study for the identification of CIR values by fitting the model to the data (Kirchsteiger et al., International Journal of Control, vol. 87, no. 7, pp. 1454-1466 , 2014). These CIR values are then used later (successfully) for glucose control. However, the proposed model was only able to describe glucose dynamics after a single meal test (around 5 hours of data).

Assuming constant, patient-specific factors for CIR and ISF is a simplification. The amount of insulin to inject per gram of CHO is actually influenced by a host of different variables, such as time of day, levels of physical activity, and psychological and physical stress.

In the present disclosure the model represented by formula (2) is used and time-dependent terms are added to allow for diurnal variations in the model parameters. Using CGM data from the clinical study in conjunction with recorded data on carbohydrate intake and insulin injections, the model parameters were identified. Against the procedures of the technique (Kirchsteiger et al., In Proceedings of the 49th IEEE conference on decision and control (CDC), 2011, pp. 5176-5181; Kirchsteiger ^{etal. ,} International Journal of Control, vol. 87, n. 7, pp. 1454-1466, 2014), direct identification procedures of the continuous time system were not used, but the parameters were determined by directly minimizing the squared error between the analytical expression for the model output in the time domain and the measured CGM data. .

The structure of model (2) is used as a basis for identifying diurnal patterns in insulin action. The structure of this model uses carbohydrates from meal D (s) and rapid intravenous insulin injections U (s) as the only inputs to the system. Carbohydrate intakes d (t) and insulin injections u (t) are represented by pulses in the time domain (with ó (t) being the Dirac delta function, K the total number of carbohydrate intakes in the frame of relevant time, Ai the mass of carbohydrates in grams of each food intake i, L the total number of insulin injections in the relevant time frame, Bj the amount of insulin in rapid intravenous injection in IU of each injection jy ti and tj corresponding times of carbohydrate intakes and insulin injections):

In previous studies (see for example, Reiterer ^{et al.,} In Prediction methods for blood glucose concentration: Design, use and evaluation, Kirchsteiger, Jorgensen, Renard and del Re, Eds., Springer International Publishing Switzerland, 2016, pp. 57-78 ), the parameters ^{K i, T i, K 2} and ^{T 2} of formula (2) have been chosen to be constant and have been adjusted in such a way as to achieve optimal alignment between the recorded CGM data and the values of BG as predicted by the model. Choosing constant values for insulin and carbohydrate sensitivity throughout the day is basically incorrect because these sensitivities are not constant factors, but change significantly depending on the time of day and other influencing factors (physical activity, injection site insulin, etc.). It is generally accepted that insulin sensitivity measured in the morning is lower than in the evening for people with diabetes. Assuming constant values for the entire period of the day means that the identified parameters correspond to averages over the entire period of the day.

To adapt to these diurnal variations, the structure of model (2) has been modified to model the effect of the time of day on insulin and carbohydrate sensitivity. In the current section, the factors ^{K i} and ^{K 2 are} no longer assumed to be constant, but a specific value is calculated for each meal intake and insulin injection. However, it is assumed that the factors change slowly and smoothly over time, so they were chosen to be described by a second-order polynomial of the time of day:

^{K u = K u K 12 • ti K 13} • ^tf (4a)

^{K 2 J = K 2 Í K 2 2 ^ l K 2 3 ^ tj} (4b)

As can be seen in (4a) and (4b), for each input a separate value of ^{K i} and ^{K 2} is calculated using the time of the corresponding carbohydrate intake (ti) and insulin injection (tj). The values are then held constant to calculate the response of the system after a specific input. It should be noted that the timing of carbohydrate intake and insulin injections can be identical, but often slightly deviates in time. In addition, there are also insulin inputs that are not linked to meals, for example correction rapid intravenous injections and carbohydrate intakes that are not linked to a rapid intravenous injection, for example when taking carbohydrates to treat or prevent hypoglycemia .

Assuming quadratic functions for both K1 and K2 turned out to be a good compromise between keeping the model simple and performing well when fitting the data. In doing so, the number of unknown parameters for the system identification problem is increased from 4 as in the original model (K1, K2, T1 and T2) to 8 (0 = [K11, K12, K13, K21, K22, K23 , T1, T2] T).

A procedure was described to identify the parameters of the model (2) using continuous time systems identification procedures (Kirchsteiger ^{et al.,} International Journal of Control, vol. 87, no. 7, pp.

1454-1466, 2014).

For the present disclosure, direct identification procedures of the continuous time system were not chosen to identify the model parameters. Instead, they took advantage of the fact that model (2) itself (together with (4a) and (4a)), as well as inputs (3a) and (3b), are described by relatively simple functions. Therefore, it is possible to perform the inverse Laplace transform for (2) and obtain an analytical expression for the integral of convolutions that describes the response of BG to carbohydrate intake and insulin injections. Therefore, the time domain model is described by (5).

In this formula, BG0 corresponds to the initial concentration of BG, t is the time (in min, with t = 0 min at the beginning of the identification time frame), and H is the Heaviside function. To derive structure (5), the BG is assumed to be in equilibrium at the beginning of an identification time frame. In the current work, these time frames were defined to start just before breakfast (around 8:00 AM) and end at 10:00 PM on the same day. Thus, an identification time frame included breakfast, lunch and dinner on the corresponding day. This is different from what was done in the prior art (Kirchsteiger et al., International Journal of Control, vol. 87, no. 7, pp. 1454-1466, 2014), where only data from breakfast were used in the system identification.

To identify the model parameters, the analytical expression of BG dynamics was used directly. Therefore, the identification of the system consisted in trying to find the optimal set of parameters that minimized the squared error between the BG calculated as described in (5) and the measured values. For each patient, the 0 parameters of an identification time frame were assumed to be constant. However, in order to limit the intrapatient variability of the model parameters, it was essential to limit the variability of the time constants T1 and T2. Therefore, the parameters for all days were determined at once and the deviation of T1 and T2 from the mean value for all days was limited to a maximum allowed value using additional constraints on optimization. Furthermore, to limit the search for optimal parameters to a physiologically relevant parameter space, minimum and maximum values have been defined for the model parameters. Therefore, the entire optimization problem is read as indicated in (6).

with: k = 1,2,3 l = 1,2 ..... N

In this optimization problem, BGmed is the vector of measured glucose values, while BGmodeo corresponds to a vector of the values calculated using equation (5) evaluated at the time points of the measurements, so the cost function is the sum of the squared errors between measured and calculated values. The output of the model in the vector BGmodel depends, of course, on the values of the model parameters Kik, i, K ² k, i, Ti, i and T ² , i with indices k (index to describe the three parameters in the quadratic equations for Ki and K2) and I (index describing each separate time frame for system identification, N time frames in total).

As already mentioned, minimum and maximum values are given for Ki, K ² , K ² / K ¹ , Ti and T ² to restrict the parameter space to a physiologically significant region. For K1 and K2 it must be ensured that the quadratic equations (4a) and (4b) remain within the search interval defined by the minimum and maximum value for Ki and K ² . This is done by imposing the first two inequalities of (6) for specific time points within each identification time frame (defined by the corresponding tmin and tmax). For current work, these inequalities are imposed at equidistant points of time separated by one-hour intervals. The same procedure was performed to impose that K / K i should be within a physiologically significant range (defined by CIRmin. And CIRmax). In Table 1 you can find a compilation of the minimum and maximum values used in the identification of the system.

In addition, a restriction was imposed to limit intrapatient variability of T 1 and T2. An inequality constraint limits the maximum deviation from the corresponding mean value (average for all N days) to At. For the current work, a maximum deviation of 25% was used for both T1 and T2 (At = 0.25). This additional restriction turned out to be necessary to be able to identify patient-specific day profiles for K2 / K1. Furthermore, the resulting K2 / K1 patterns showed a much better similarity to the diurnal variations of the CIR factor used during the clinical trial to calculate the need for rapid intravenous insulin injection.

In addition, restrictions have been introduced to limit the intrapatient variability of the K1 and K2 profiles to a reasonable value. The variations between days of the K1 and K2 profiles were limited to 30% (Ak = 0.30), which were assumed to be reasonable values (see, for example, Heinemann, Diabetes Technology & Therapeutics, vol.

4, no. 5, pp. 673-682, 2002) for intrapatient variability of insulin absorption and action).

In addition, the K ² values were imposed so that they did not vary too much from some predefined reference ISF values.

In one example, the variability between days of T1 and T2 was restricted to 25% (At = 0.25) and the variations between days of the profiles of K1 and K2 were limited to 30%. (Ak = 0.30), which were assumed to be reasonable values (see, for example, Heinemann, Diabetes Technology & Therapeutics, vol. 4, no. 5, pp. 673-682, 2002) for intrapatient variability of absorption and action of insulin).

By simply rearranging the terms, the inequality constraints in (6) can be rewritten in the form A • 0 <b (where 0 corresponds to the vector of parameters containing the unknown parameters of the model), which means that the problem reduces to an optimization problem (not convex) subject to linear inequalities. Therefore, this optimization problem can be solved using most of the standard optimization tools. The MATLAB routine FMINCON was used to identify the values for the parameters.

Results

The procedure described above has been applied to the clinical data of the trial (Zschornack et al., Journal of Diabetes Science and Technology, vol. 7, no. 4, pp. 815-823, 2013). For each day, the CGM device data with the lowest MARD for that specific day was chosen. However, the SMBG could also have been used for system identification without any modification to the methodology, but the SMBG data, of course, has a much coarser time resolution. For simplicity, it is assumed that the CGM devices used in this assay are perfectly calibrated and read transient values from the BG without additional time delays. This assumption is a simplification that has been made throughout this thesis. However, since the measurement efficiency of the CGM system used in the test is very high (MARD less than 10%) and the time delay of the specific CGM device is very short, this assumption seems to be justified for the whole of data used in this work. Alternatively, CGM signals could be post-processed in conjunction with SMBG to account for loss of CGM calibration, as well as lag times between BG and interstitial glucose.

The identification of the system has been carried out for the entire patient database and using the procedure described above. As a result, reasonable accommodation could be obtained for all patients and almost every day of the clinical trial. To evaluate the quality of the identified models, the FIT value (yr model output, yr measurement, y: mean over measurements, Ntot: total number of measurements) was used:

FIT - 100

Table 1

After careful analysis of the system identification results, it was assumed that the general dynamics of BG could be described reasonably well for the case of FIT values greater than 40% (technical criterion). Therefore, 40% was used as a threshold to distinguish between good and bad descriptions of BG dynamics and only models with a FIT value above this threshold value were considered in subsequent analyzes. Notably, sensitivity calculations with different FIT cutoff values (even as low as 10%) resulted in very similar results when comparing calculated K2 / K1 values with clinically used CIRs (discussed later in this section). , which shows the robustness of the procedure. In 75% of cases, a model with a FIT value above the threshold could be identified using the methodology described above. For 25% of the cases with a FIT value below the threshold, the poor description of BG dynamics is probably due to the strong impact of some non-modeled disturbances (for example, physical or psychological stress).

In figs. 1 to 6 you can see an example of the model identified for a specific patient (patient 0013) and all the full days of the trial (days 2 to 7).

The first panel of each figure shows CGM (as well as SMBG) data from the clinical trial along with glucose dynamics as described in the fitted model. For this case, a reasonable agreement between the measurements and the model output could be achieved for all days (FIT values between 63.81% and 69.93%). In the second panel you can see the carbohydrate intakes and insulin injections recorded during the test. These were used as inputs to the model. The third panel shows the evolution of K1 and K2 as described in equations (4a) and (4b) with the values of the identified model parameters, as well as the resulting diurnal K2 / K1 changes. It should be noted that K1 is positive, while K2 is negative, and that the numerical values of K1 are approximately one order of magnitude smaller than that of K2.

By combining the evolution of K2 / K1 against the time of day over several days in a graph, a representation with interesting patient-specific information can be generated. Comparing diurnal changes over several days, it is possible to draw conclusions as to whether or not a patient has a characteristic profile for K2 / K1 that appears more or less the same for every day or whether there are larger changes in conformation from day to day. In addition, the distance between the curves can be used as a measure of intrapatient variability of insulin action. These types of characteristic features could be used to classify patients into different groups for which it would be better to seek a specific strategy for managing BG.

In figs. 7, 9 and 10 you can see examples of this type of graph for three different patients (0013, 0033 and 0051). In these graphs, the diurnal K2 / K1 profiles for each data day are shown in blue (for FIT> 40%) or green (for FIT <40%), while the average profile corresponds to a bold black line. For patient 0033 (Fig. 9) it was found, for example, that the typical pattern for K2 / K1 is relatively stable as a function of time of day (with a slight maximum around lunchtime) and shows a degree moderate intrapatient variability that appears to be maximum around the middle of the day.

However, a typical conformation could not be identified for all patients. In fig. 7 An example of a patient with a more pronounced inter-day variability of profiles can be seen for patient 0013. For that patient, four curves have a maximum of K2 / K1 near lunchtime, while one curve monotonically decreases from breakfast until dinner. Patient 0051 in FIG. 10, on the other hand, shows a consistent conformation of the K2 / K1 profiles, but a large distance between the curves for each day, indicating a high level of intrapatient variability.

In addition to information on intrapatient K2 / K1 variability and inter-day consistency of relative conformations, the K2 / K1 curves from the system identification can also be compared to the CIRs used by the patients in the trial to calculate needs. insulin as a rapid intravenous injection. Of course, the same can also be done for K2 profiles (see Fig. 8 for such an example), which can be compared to clinically used ISF values. Clinically used CIR and ISF factors are already included in Figs. 7 to 10 as red circles. In case these values have changed during the course of the clinical trial, the old values are still included in the graphs as white circles. Specific CIR values for breakfast, lunch and dinner have been used in the clinical trial. It can be seen from the four illustrative figures that there is a fairly good agreement between the identified curves and the factors used clinically. Only for the case represented in Fig. 10, where there are strong variations between days for the K2 / K1 curves, the distance between the average diurnal profile (bold black line) and the CIRs used clinically is slightly larger.

A comparison of K2 / K1 was made with the corresponding CIR values for all patients and every day and the good correlation between K2 / K1 and CIR could be confirmed. The same is true for K2 and clinically used ISF values. This comparison can be seen in fig. 11a and 11b. Calculating the proportion of the two values ((-K2 / K1) Calculated / CLINICAL CIRCUIT and (-K2) Calculated / ISFClinical) for all patients, all meals and all days of the trial results in a distribution with a median of 1.06, quartile of 25% of 0.94 and quartile of 75% of 1.21 for CIR and with a median of 0.87, quartile of 25% of 0.74 and quartile of 75% of 1.01 for ISF (see Fig. 11b and Table 1). It can be seen that for the ISF values there is a systematic negative bias (the clinically used ISF values are higher than those identified from the data). This could be a deliberate overestimation of the ISF values used clinically to avoid hypoglycemia after correction rapid intravenous injections. For the CIR values, no significant bias was found between the calculated values and those used clinically.

In figs. 12 to 14, additionally, the comparison between the calculated CIR and ISF values and those used clinically for each of the meal times (breakfast 8:00 am, lunch 1:00 pm, dinner 6:00 pm) are shown separately. Through this more detailed analysis, it can be seen, for example, that the bias between calculated and clinically used ISF values is minimal at breakfast and increases toward dinner. Additionally, it is visible for the CIR values that the correlation between the calculated values and those used clinically is minimal for lunch time, while for both breakfast and dinner, a very high correlation can be seen. For none of the meals is a strong systematic bias in CIR.

Another interesting result is found when analyzing those cases in which the CIR value was modified during the course of the study. Comparing, for example, the difference for the CIR at breakfast with the corresponding mean value of K2 / K1 for patient 0051 (see Fig. 10), it was found that this difference decreases significantly with the change in CIR, that is, the new value tends to be closer to the calculated K2 / K1 than the old one. When analyzing all the cases in which the CIR was changed during the trial combined (all values for all patients and all meals combined), a reduction in the mean difference between the clinically used CIR and the K2 / K1 identified in 7.9% (from 22.2% to 13.3%). This again is an indication that the K2 / K1 and K2 profiles from the system identification are related to the patient-specific "true" CIR and ISF values.

However, it should not be forgotten throughout the analysis that the patient and mealtime specific CIR and ISF values used by the patients in the trial (Zschornack et al., Journal of Diabetes Science and Technology, vol. 7, no. 4, pp. 815-823, 2013) have not been systematically verified as part of the clinical protocol, but have only been modified if the clinical staff conducting the clinical study have identified inappropriate settings. Therefore, clinically used settings do not necessarily correspond to the best possible and the difference between identified and clinically used values does not necessarily indicate a failure of the system identification approach. However, having clinically calculated and used values close to each other is an indication of the good quality of the recently proposed methodology.

Based on the results presented above, it was found that the proposed procedure is adequate for the identification of CIR and ISF values from the recorded data. Such an approach may be useful, for example, to guide patients with DM1, but also healthcare professionals with less experience in evaluating diabetic diary books and fine-tuning the BC settings. . Today, as insulin pumps and CGM devices are increasingly widespread among DM1 patients and newer devices are also increasingly interconnected, it seems feasible to collect the data required for such a system identification algorithm. automatically routinely. This is even more important considering that newer devices have smartphone interfaces and allow uploading of collected data. Additionally, today many patients with DM1 use diabetes diary log applications that facilitate the collection and integration of food data (which can hardly be collected automatically). Automatic calculation of values CIR and ISF seems therefore feasible with minimal extra effort.

It is worth mentioning that there is not only a variability between days in glucose dynamics. Furthermore, glucose dynamics can also change over time if patients change their lifestyle and, for example, experience a change in insulin resistance. The recently proposed procedure could be used to track such changes and adjust CIR and ISF values.

Based on this reasoning, the following procedure is proposed. Such a procedure or methodology may also be referred to as an adapted rapid intravenous injection (ABC) calculation. For example, the characteristics of this procedure can be summarized as follows:

- CGM data is collected with information on rapid intravenous injections and food intake for N days.

- The methodology described above is applied to this data set and the diurnal profiles for K2 / K1 and K2 are identified.

- Combined profiles for -K2 / K1 and -K2 can be used to calculate a profile for CIR and ISF as a function of time of day. In the simplest case, only the average profile (average of the identified profiles for each day) is used.

- Based on the identified values of T2, a value for the DIA can be estimated.

- CIR, ISF and DIA values can be applied in a standard BC using formula (1).

- As soon as data for the next day is available, it can be added to expand the size of the data set or replace data from the oldest day. Using this modified data set, the system identification can be repeated by providing new values for CIR and ISF as a function of time of day. These can override the current setting on the patient's BC.

- Alternatively, the cost function in the optimization problem (see (6) above) can be altered to give newer data more weight.

- The resulting algorithm should be able to track changes in the patient's rapid intravenous insulin needs that could occur over a longer period of time (for example, CIR decreases over time due to a change in insulin resistance insulin related to an increase in the patient's body weight).

Other examples

Until now, ABC has always been implemented and applied so that data from multiple days is collected and adjusted using the procedure described above. A further discussion is provided below regarding how much information is needed to identify suitable CIR and ISF values using the ABC approach. To analyze this question, a test calculation is performed using the ABC procedure as follows:

1. During the first full day of the trial (day 2), patients use their standard BC settings (BC settings agreed with the healthcare professional) to calculate their rapid intravenous injection needs. During this day, data is collected for the ABC.

2. The data from day 2 are fitted and the resulting CIR and ISF profiles are used on day 3 to calculate the need for rapid intravenous injection (together with the estimated value of T2 to calculate the IOB). The corresponding glucose trajectories are calculated in the drift analyzes.

3. Based on the data from the first two days, the ABC methodology is used to identify the CIR and ISF profiles. The corresponding average profiles are used to calculate the need for rapid intravenous injection on day 4 (in conjunction with the average T2 to calculate the IOB).

4. Stage 3 is repeated for days 5, 6 and 7. The corresponding database for identification with the ABC strategy is continually expanded during this time.

Additionally, as a reference scenario, the case of calculating the needs for rapid intravenous injection was also analyzed with the CIR and ISF values used clinically. This configuration will be renamed StdBC.

Glucose trajectories resulting from drift analyzes are analyzed for time throughout the day (from 8:00 a.m. to 10:00 p.m., or the predefined or determined time that the patient remains awake) and with respect to the percentage of time in hypoglycemia, the percentage of time in hyperglycemia, as well as the value of the cost function V (see (10) below) with W = 5. The resulting mean values for the 37 patients combined can be seen in figs. 15 to 17.

When analyzing those graphs, it becomes clear that the behavior of ABC and StdBC as a function of the study day is quite similar. The curves for time in hypoglycemia, time in hyperglycemia and cost function V are almost parallel to each other with a more or less fixed distance. During the course of the study, the time in hypoglycemia and hyperglycemia, as well as V, continuously decrease (the only exception is for the time in hyperglycemia on day 5). This general behavior can be explained by a run-in period during the start of the trial. The patients are in an unfamiliar environment that causes an alteration in their daily routines and most likely results in poorer BG control in the initial phase of the trial. Interestingly, after just one day of data, AUC leads to a shorter hypoglycemic time and a lower V value than StdBC. Time in hyperglycemia is comparable for ABC and StdBC, but slightly higher for ABC every day. Based on these findings, it can be concluded that a single day of data is basically enough for ABC to perform favorably. Using a larger data set, with more days of course, reduces the effect of possible outliers, for example, days of data that can hardly be explained by the model in the ABC procedure or days with low-quality data. Additionally, it should also be noted that the best strategy is not necessarily to continue to simply add data for the identification of the ABC system and then calculate the average CIR and ISF. Instead, it might make more sense to put more weight on getting newer data sets or discarding older data sets from analysis.

In addition to the run-in effect of ABC, an online application of ABC was also studied in which information from previous meals on the same days is used to adjust the dosing of the rapid intravenous injection of subsequent meals. In the baseline ABC, the data for full days is adjusted and analyzed together. However, this does not necessarily have to be done this way. For example, it may happen that the amount of rapid intravenous injection for a breakfast calculated with the CIR and the ISF of the ABC leads to hypoglycemia. This would indicate that the amount of rapid intravenous injection was too high for the food ingested and, assuming the carbohydrate estimate was accurate, that the CIR and / or ISF were inappropriate.

The question then is how to adjust CIR and ISF at lunchtime to get better performance for that meal.

First, the question about the correlation between CIR and ISF values between different meals on the same day must be analyzed. In the absence of such a correlation, it would not make sense to adjust the CIR for lunch time if the CIR applied for breakfast were too low. In fig. 18, all the identified CIR values from different meals for all patients and all days are plotted against each other, while in fig. 19 the same is done for ISF values. The values are normalized by the mean value of the corresponding patient to make them comparable between patients. It can be seen that there is no strong and obvious correlation between CIR values for contiguous meals. The linear regression, however, reveals that there is a positive correlation between the CIR values for breakfast and lunch, as well as between those for lunch and dinner. Between the breakfast and dinner factors, on the other hand, there is no correlation. The same happens with the ISF values (a strong correlation was even found between the ISF values for lunch and dinner). This would indicate, for the example above, that after a breakfast with too much insulin applied, it might be advantageous to increase the CIR and ISF values to calculate the rapid intravenous injection needs for the next lunch.

To do this, the following procedure is suggested:

1. Full-day data is used to identify the CIR and ISF baseline profiles using the ABC methodology described above.

2. The CIR and ISF values suggested by ABC are used to calculate the rapid intravenous injection needs for breakfast and the corresponding amount of insulin is injected.

3. The glucose trajectory after breakfast (from the start of the meal to just before lunch time) is included in the AUC data set and adjusted together with the data from the previous days to obtain an estimate. of CIR and ISF suitable for breakfast eaten.

<¡rzo) para el próximo almuerzo se ajustan a continuación de acuerdo con la siguiente fórmula: 4. The values of CIR (C / fi "'^ ¡rzo) and

<Rzo) for the next lunch are then adjusted according to the following formula:

where CIR ^{2 l} ¿, ^¿ and ISF ^% , for the calculation of the adapted rapid intravenous injection (AUC), the CIR and the ISF for lunch time from the daytime average profile of the AUC (calculation of the adapted rapid intravenous injection), and where ClR ^{^} H ^{^ yuno} and ClR ^{^} H ^{^ yuno are} the corresponding values for breakfast of the average diurnal profile. ClR ^ ^l ^ 110 and ClR ^ ^l ^ 110 , on the other hand, are the CIR and ISF factors identified from the postprandial glucose trajectory from breakfast eaten earlier on the same day. The factor f is an adjustment parameter. In case it is set to 0, the information from the previous breakfast on the same day will be ignored, while setting it to 1 assumes that the same relative deviation will occur at lunch time compared to the average daytime profile observed for breakfast.

5. The current CIR and ISF values are used to calculate the dose of rapid intravenous injection of insulin for lunch to be injected next.

6. Just before dinner, the data for the same day are adjusted using the ABC methodology to obtain estimates for the CIR and ISF values that would have been appropriate for lunch ( CIR ^{Il ^ f ¡^ erzo} and CIR ^f ™ ^£ f ^rzo ). The CIR and ISF for dinner are then adjusted analogously to the procedure for correcting for lunchtime factors:

f j p c e n a i m zo rir> aBlmCoffee '

^ inac tu al _ - ^r l ⁱ i ^D k ^c a ^e b ⁿ c ^{a L} r ^t Da ^K Slys¡uder ^L l ^K A

{if njr> lunch ^LlH ABC (9.1)

jrrcena ^{- jcucena} jcnlunch tcualmuerzox ^{l ^ t} Sys¡d ^{- l ¿t} ABC \

l D rac tu al - _{l ^^ ABC} , _{{! ! •} jcn & lw-uerzo ¹ (9.2)

^lst ABC J

This procedure has been tested using the data (cf. Zschornack et al., Evaluation of the performance of a novel system for continuous glucose monitoring, Journal of Diabetes Science and Technology, vol. 7, no. 4, pp. 815-823 , 2013) for different factors f (between 0 and 2). Days 2 to 6 have been used to identify an average daytime profile, while on day 7 the strategy indicated above is applied. The results have been analyzed with respect to the percentage of time in hypoglycemia, the percentage of time in hyperglycemia, as well as with respect to the value of the cost function V (with W = 5):

V - thiper / ttot + 'W' thipo / ttot (10)

V has been defined as the weighted sum (with weighting factor W) of the time in hypoglycemia (thipo) and hyperglycemia (thiper), both normalized by the total time considered for the performance evaluation (ttot).

The results of this evaluation are plotted as a function of f in Figs. 20 and 21. It can be seen that, for f between 0 and 1, the time in hypoglycemia and the cost function V do not change much, but they begin to increase rapidly for values of f greater than 1. The time in hyperglycemia, on the other hand On the other hand, it increases almost linearly between f = 0 and f = 2. The lowest value of V is found for f = 0.3. This is due to a slight decrease in the time in hypoglycemia (from 1.69% to 1.48%).

The characteristic features disclosed in this specification, in the figures and / or in the claims may be important for the realization of various embodiments, taken in isolation or in various combinations thereof.

Claims (7)

A method for determining a rapid intravenous injection to be administered by an insulin delivery device, comprising, in a data processing device having one or more processors, - receiving glucose measurement data indicative of glucose level measurement signals detected by a continuous glucose sensor for a person; and - determining, by the one or more processors at a point in time, for the person, a rapid intravenous injection of insulin to be administered by the insulin delivery device, determining the rapid intravenous injection of insulin based on a functional relationship provided by a formula comprising formula expressions indicative of - a glucose level derived from glucose measurement data; - a ratio of carbohydrates to insulin (CIR); and - an insulin sensitivity factor (ISF); wherein the determination further comprises at least one of, - for the carbohydrate to insulin ratio, determine a current CIR value based on CIR data indicative of the carbohydrate to insulin ratio for at least one of - a period of time from the start of a previous meal until before the start of a current meal on a current day comprising the point in time; - current food; and - the previous meal; and - for the insulin sensitivity factor, determine a current ISF value based on ISF data indicative of the insulin sensitivity factor for at least one of - the period of time from the start of the previous meal to before the start of the current meal on a current day comprising the point in time; - current food; and - the previous meal; wherein the determination of the rapid intravenous injection of insulin is carried out by a data processing device provided in a system selected from the following group: an insulin pen; an insulin pump; a remote control device configured to remotely control an insulin pump or insulin pen; a continuous glucose meter; a controller for a continuous glucose meter; a blood glucose meter; and a portable device such as a mobile phone, tablet or smart watch, characterized in that the determination of rapid intravenous insulin injection further comprises at least one of - determine a rapid intravenous injection of insulin for lunch after a previous breakfast eaten on the current day and - determine a rapid intravenous injection of insulin for dinner after a previous lunch eaten on the current day, where, for lunch, the current CIR value ^{(CIR ^ Z ^ a} ™ 0) and the current ISF value ^{(ISF ac ^} frzo) are determined as follows:

in which - CIR £ jJ¡lunch and ISF% g ™ exertion are, for an ABC adapted rapid intravenous injection calculation, the carbohydrate-to-insulin ratio and insulin sensitivity values for lunch derived from an insulin level profile. average daytime glucose for the person; - ClR ^ lSg Vuno and lSF ^ ggayuno are the carbohydrate-to-insulin ratio and insulin sensitivity values for breakfast derived from the person's average daytime glucose level profile; - ClRgy ^ g ^ uno and lSFfySs [gyuno are the carbohydrate to insulin ratio and insulin sensitivity values determined from a breakfast postprandial glucose trajectory from the previous breakfast eaten earlier in the week. current day for the person; and - the factor f is an adjustment parameter; and where, for supper, the current CIR value ( C IR ^ icu) and the current ISF value ( ISF ^ audi) are determined as follows:

in which - CIRab ™ and ISFab ™ are, for the ABC adapted rapid intravenous injection calculation, the carbohydrate to insulin ratio and insulin sensitivity values for dinner derived from the average daytime glucose level profile for the person; - CIR '^ Effort and ISFJ ^ j ^ uence are the carbohydrate to insulin ratio and insulin sensitivity values for lunch derived from the average daytime glucose level profile for the person; - CIRsy ™, “lerzo and ISFs' ™% erz ° are the carbohydrate to insulin ratio and insulin sensitivity values determined from a lunch meal postprandial glucose trajectory previously ingested earlier in the current day for the person; and the factor f is a fitting parameter, The method of claim 1, wherein the determination comprises determining the rapid intravenous injection of insulin based on the functional relationship provided by the formula further comprising formula expressions indicative of the total amount of insulin delivered within a first time interval; and - the total amount of carbohydrates ingested by the person of interest within a second time interval. The method according to at least one of the preceding claims, further comprising at least one of - sending rapid intravenous injection data indicative of the rapid intravenous injection of insulin to be delivered by the insulin delivery device through an output device connected to the data processing device; and - transmitting the output data to the insulin delivery device. 4. The method according to at least one of the preceding claims, wherein the Rapid intravenous injection of insulin is determined based on the following formula: CHO G r _{Rap reu} intravenous injection ida _{p ob jet iv o} + - IO B CIR Tsf where CHO is the carbohydrate content of a meal in grams, Gpre is the preprandial glucose level, Goject is a target glucose level for the postprandial glucose level, ISF is the insulin sensitivity value, CIR is the carbohydrate value relative to insulin, and IOB is insulin on board. Computer program product, comprising a program code configured to, when loaded into a computer having one or more processors, perform the method according to at least one of the preceding claims. 6. Data processing device, comprising one or more processors configured to determine a rapid intravenous injection to be administered by an insulin pump by a method of at least one of claims 1 to 4. The data processing device of claim 6, wherein the one or more processors are provided in a system selected from the following group: an insulin pump; an insulin pen; a remote control device configured to remotely control an insulin pump or insulin pen; a continuous glucose meter; a continuous blood glucose meter; a controller for a continuous glucose meter; a controller for a continuous blood glucose meter; and a portable device such as a mobile phone, a tablet, or a smart watch.